COVID-19 is a respiratory disease that directly affects the lungs of infected individuals, leading to severe respiratory problems and lung infections. Although the severity of COVID-19 has decreased, the possibility of contracting the virus still exists, especially for individuals with underlying medical issues. Diagnosis of the severity of COVID-19 is very important in providing essential services to patients, improving treatment outcomes, and reducing complications and mortality rates associated with the virus. But, distinguishing of the severity of COVID-19 is a challenging task. COVID-19 is divided into four classes as far as its severity is concerned: normal-PCR+, mild, moderate, and severe. To overcome this challenge, we have introduced a novel method called DA-COVSGNet, based on a Convolutional Neural network (CNN). In The proposed model, we preprocessed X-ray images using Synthetic Minority Over-Sampling Technique (SMOTE) and Contrast Limited Adaptive Histogram Equalization (CLAHE) techniques, and fed them to the CNN architecture. Furthermore, we used new attention mechanisms to aid in better distinguishing and classifying disease severity levels, resulting in higher accuracy in classifying disease severity classes. Finally, we evaluated our proposed method on the COVIDGR dataset. The results show that our proposed method achieved accuracies of 96.7%, 96.2%, 98.5%, and 95.2% for the categories of Normal-PCR+, mild, moderate, and severe, respectively.

In deep reinforcement learning, experience replay buffers are used to reduce the effects of sequential data and make better use of past experiences. Prioritized Experience Replay (PER) improves upon random sampling by selecting transitions based on their temporal difference (TD) error. However, PER does not consider how important each transition is or how many times it has been used during training. In this paper, we propose a new method for adaptive prioritization that takes into account three additional transition-level factors: reward, usage count (counter), and policy probability—collectively referred to as RCP values. These values are normalized and used alongside the TD error to calculate the probability of selecting each transition from the replay buffer. We evaluate our method on several Atari environments and show that using any of the RCP values individually can improve performance compared to standard PER. To combine all three RCP components, we explore three aggregation functions: minimum, maximum, and mean. Experimental results show that the best aggregation method depends on the environment. However, the mean function generally provides stable improvements across tasks, as it balances all RCP signals and avoids over-relying on any single factor.

Elastic optical network has recently raised immense research efforts to serve as a highlyflexible and scalable platform required by the emerging on-demand cloud applications. Besides, virtual network embedding on top of elastic optical network has been introduced as a promising solution to fine-grain sharing of node processing and optical bandwidth resources. Owing to the diverse delay sensitivity requirements of heterogeneous cloud applications, an efficient double queue scheduler is presented, to deliver differentiated setup delay embedding in a fair manner. Extensive numerical simulations carried out over two prototype substrate networks, demonstrated that our proposed scheme can concomitantly enhance blocking and fairness performance of the single queue scheduler over wide range of offered traffic loads.

In this paper, we investigate the usage of machine learning in the detection and recognition of double stars. To do this, numerous images including one star and double stars are simulated. Then, 100 terms of Zernike expansion with random coefficients are considered as aberrations to impose on the aforementioned images. Also, a telescope with a specific aperture is simulated. In this work, two kinds of intensity are used, one is in-focus and the other is out-of-focus of the telescope. After these simulations, a convolutional neural network (CNN) is configured and designed and its input is simulated intensity patterns. After learning the network, we could recognize double stars at severe turbulence without needing phase correction with a very high accuracy level of more than 98%.

This study presents a new procedure based on Artificial Neural network (ANN) for identification of double cracks in Functionally Graded Beams (FGBs). A cantilever beam is modeled using Finite Element Method (FEM) for analyzing a double-cracked FGB and evaluation of its first four natural frequencies for different cracks depths and locations. The obtained FEM results are verified against available references. Furthermore, four Multi-Layer Perceptron (MLP) neural networks are employed for identification of locations and depths of both cracks of FGB. Back-Error Propagation (BEP) method is used to train the ANNs. The accuracy of predicted results shows that the proposed procedure is suitable for double cracks identification detection in FGBs.